Skip to content

docs: observability page facts (issue #83) - #90

Open
SebConejo wants to merge 1 commit into
mainfrom
fix/issue-83-observability
Open

docs: observability page facts (issue #83)#90
SebConejo wants to merge 1 commit into
mainfrom
fix/issue-83-observability

Conversation

@SebConejo

Copy link
Copy Markdown
Member

Processes the eight findings of #83, recommended options, each re-verified in the platform code, plus a live check of the usage endpoints.

Finding 1 (option 1) — the limit period is not rolling. It is a calendar window on the server's clock: day at midnight, week on Monday, month on the 1st, hour at the top of the previous clock hour, restart at the next boundary. The page now agrees with the gateway page's "the block clears at the start of the next period". Deliberately not UTC: the boundaries are computed in the server's local timezone.

Finding 2 (option 1) — a failed attempt does not burn tokens. The recorder writes cost and tokens on the success paths only; failure rows keep zero tokens and no cost. The paragraph now says what the record holds instead of the wrong justification.

Finding 3 (option 1) — the Tip claimed Manifest cannot price custom-provider models. The connect form asks for the two prices and Manifest computes cost from them. The Tip now states the conditional truth, and providers/custom-providers.mdx carried the same wrong claim, fixed in the same PR as the issue prescribed.

Finding 4 (option 1) — the crossing is recorded before the email is sent and never retried inside the period, so a failed send costs the period its email while Triggered still counts up. One sentence on the evaluation paragraph, and the self-hosted email block now points at Send test email.

Finding 5 (option 1) — Pending and Cancelled join the request model, and one paragraph explains why the Overview's Failed requests card links to a longer list than the number on the card: the Overview counts finished requests, the Failed filter returns everything that did not succeed.

Finding 6 (option 1) — the usage API section now carries the range parameter with its accepted values, the two different defaults (24h versus 7d), the rolling-versus-calendar warning, and one sample response per endpoint. I checked the shapes live on a local instance: /usage answers 24h by default with agentName, /costs answers 7d without it, and range=1d gets a 400. The sample numbers are illustrative; the field lists are the live ones.

Finding 7 (option 1) — the Pro lock on long ranges applies on the Overview and on the Requests log, and the log's All time option stays free. The "covers up to 7 days of history" phrasing is gone, since it read as a retention limit.

Finding 8 (option 1) — the Limits card said "plus email provider setup" to every reader, while the picker only exists on self-hosted. The card now reads "Threshold rules per harness." and the self-hosted block keeps the setup instructions.

Coherence GREEN (48 pages, 57 vars).

Closes #83

…ts, custom-provider pricing, alert delivery, request states, usage API shapes, Pro ranges, Limits card (issue #83)
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Docs audit — observability (requests page, analytics) — 2026-08-19

1 participant